Glucose–insulin models
Glucose–insulin models describe how the body regulates glucose concentration through interacting physiological processes. They are classic examples of mathematical feedback systems.
The feedback loop
Once glucose falls, the stimulus for insulin secretion also weakens. The two variables therefore influence each other through time.
Choose the variables
Let
\[G(t)=\text{glucose concentration},\qquad I(t)=\text{insulin concentration}.\]A general model can be written
\[\boxed{\frac{dG}{dt}=\text{glucose input}-\text{glucose removal},}\]\[\boxed{\frac{dI}{dt}=\text{insulin secretion}-\text{insulin clearance}.}\]This balance structure is more important than any particular choice of formula.
A simple glucose equation
One teaching model is
\[\boxed{\frac{dG}{dt}=R(t)-a(G-G_b)-bI(G-G_b).}\]Here \(R(t)\) is an external glucose input, \(G_b\) is a baseline glucose level, \(a\) represents insulin-independent return toward baseline, and \(bI(G-G_b)\) represents insulin-dependent glucose removal.
A simple insulin equation
Insulin secretion can be modelled as increasing when glucose rises above baseline:
\[\boxed{\frac{dI}{dt}=s\,[G-G_b]_+-kI,}\]where
\[[x]_+=\max(x,0).\]The first term represents glucose-stimulated insulin secretion and \(kI\) represents insulin clearance.
Why use deviations from baseline?
It is often convenient to define
\[g=G-G_b,\qquad i=I-I_b.\]Then \(g=0\) and \(i=0\) represent the baseline state. This makes the mathematical effect of a perturbation easier to see.
A linear feedback model
Near a physiological operating point, a nonlinear system can sometimes be approximated by
\[\boxed{\frac{dg}{dt}=-ag-bi+R(t),}\]\[\boxed{\frac{di}{dt}=cg-di.}\]Glucose stimulates insulin through \(cg\), while insulin reduces glucose through \(-bi\).
What happens after a glucose disturbance?
Suppose glucose is suddenly above baseline while insulin initially remains near baseline. The positive glucose deviation stimulates insulin. As insulin rises, the negative insulin term in the glucose equation becomes stronger, helping drive glucose downward.
The response does not have to be instantaneous because insulin itself is a dynamical variable.
Model response to a glucose pulse
Why insulin peaks after glucose
Insulin is stimulated by elevated glucose but requires time to accumulate. Consequently, glucose can begin high while insulin rises later.
This lag is a consequence of the coupled differential equations and does not require drawing an assumed delayed curve.
Homeostasis
Homeostasis means maintaining physiological variables near regulated ranges despite disturbances.
Mathematically, a baseline state can be represented by an equilibrium. If there is no external glucose input, an equilibrium satisfies
\[\frac{dG}{dt}=0,\qquad\frac{dI}{dt}=0.\]After a small disturbance, a stable equilibrium attracts the system back toward baseline.
Stability of the linear model
The matrix form is
\[\frac{d}{dt}\begin{pmatrix}g\\i\end{pmatrix}=\begin{pmatrix}-a&-b\\c&-d\end{pmatrix}\begin{pmatrix}g\\i\end{pmatrix}.\]The Jacobian is therefore
\[J=\begin{pmatrix}-a&-b\\c&-d\end{pmatrix}.\]Its trace and determinant are
\[\operatorname{tr}(J)=-(a+d)<0,\qquad \det(J)=ad+bc>0\]when \(a,b,c,d>0\). These conditions support local stability of the baseline equilibrium.
Insulin sensitivity
Insulin sensitivity describes how strongly insulin affects glucose handling. In the simple model, the parameter \(b\) controls the strength of insulin-dependent glucose removal.
A larger \(b\) means a given insulin level has a stronger modelled effect on glucose dynamics.
Compare different insulin sensitivities
Insulin resistance in a model
Reduced insulin sensitivity can be represented by decreasing the parameter controlling insulin action.
Then a given insulin concentration produces less glucose removal, so the feedback system may require a larger or more prolonged insulin response to control the same glucose disturbance.
Glucose input from meals
A meal does not usually place glucose into blood instantaneously. A simple absorption compartment \(Q(t)\) can be introduced:
\[\frac{dQ}{dt}=-k_aQ,\]with glucose appearance
\[\boxed{R(t)=fk_aQ(t).}\]The input therefore rises or falls according to an explicitly defined absorption process rather than an arbitrary curve.
Meal absorption function
If \(Q(0)=Q_0\), then
\[Q(t)=Q_0e^{-k_at}\]and
\[\boxed{R(t)=fk_aQ_0e^{-k_at}.}\]More detailed gastrointestinal models can include several transit compartments to create delayed and broader glucose appearance profiles.
Pancreatic insulin secretion
Insulin secretion is nonlinear in reality. A saturating teaching function might be
\[\boxed{S(G)=S_{max}\frac{(G-G_0)_+^n}{K^n+(G-G_0)_+^n}.}\]This represents little secretion below a threshold scale, increasing secretion as glucose rises, and saturation at high glucose.
Glucose-dependent secretion curve
Basal insulin and stimulated insulin
Models may separate a basal secretion rate from glucose-stimulated secretion:
\[\frac{dI}{dt}=S_b+S(G)-k_I I.\]At baseline, secretion and clearance balance.
The minimal model idea
A well-known modelling approach for glucose-tolerance data introduces an additional variable representing insulin action rather than assuming measured insulin acts on glucose instantaneously.
A schematic form is
\[\frac{dG}{dt}=-p_1(G-G_b)-XG+R(t),\]\[\frac{dX}{dt}=-p_2X+p_3(I-I_b).\]The variable \(X(t)\) represents delayed insulin action on glucose disposal.
Why introduce an insulin-action variable?
Measured plasma insulin and its physiological effect need not change at exactly the same rate. Insulin must signal through tissues before its full glucose-lowering action is expressed.
The state \(X(t)\) gives the model its own timescale for this effect.
Glucose effectiveness
Glucose can influence its own disappearance independently of changes in insulin. In minimal-model language, this motivates a parameter representing glucose effectiveness.
Mathematically, it corresponds to glucose-dependent return toward baseline that does not require the insulin-action variable.
Oral versus intravenous glucose models
After intravenous glucose, input can be represented as an initial condition or known infusion. After oral glucose, gastrointestinal absorption must also be represented.
The same measured glucose curve can therefore arise from different combinations of absorption, insulin secretion and insulin sensitivity.
Why glucose data alone may be insufficient
If only glucose is measured, slow glucose decline could result from weak insulin secretion, weak insulin action, altered glucose production or slow absorption.
Simultaneous glucose and insulin measurements help distinguish these mechanisms, although identifiability can still be difficult.
Liver glucose production
The liver contributes glucose through endogenous production. Insulin can suppress this production, while other hormonal signals can increase it.
A more detailed balance is therefore
\[\frac{dG}{dt}=\text{gut input}+\text{hepatic production}-\text{glucose utilisation}.\]Other hormones
Glucagon, incretins, cortisol, catecholamines and other signals can affect glucose regulation. They can be added as state variables or represented through effective terms when required by the modelling question.
Type 1 diabetes as a modelling change
A model of severe loss of endogenous insulin secretion can reduce or remove the pancreatic secretion term and introduce exogenous insulin as an input.
The resulting system can be used to study how external insulin interacts with glucose dynamics.
Type 2 diabetes as a modelling change
Models may represent combinations of reduced insulin sensitivity, altered insulin secretion and changes in hepatic glucose regulation.
Because several mechanisms can produce similar glucose observations, parameter interpretation requires suitable data and careful model validation.
Exogenous insulin
Injected insulin does not act instantaneously. A pharmacokinetic submodel can describe absorption and plasma insulin, which then feeds into an insulin-action model.
Continuous glucose monitoring
Continuous glucose monitors provide dense time-series data, but the measured interstitial glucose signal can differ from blood glucose and can contain delay and measurement noise.
A mathematical observation model can distinguish the underlying physiological state from the measured signal.
Stochastic glucose models
Meal timing, absorption, physical activity, hormonal responses and measurement error vary between occasions. Stochastic models can represent some of this variability explicitly.
The goal is then not only to predict a mean trajectory but also to describe uncertainty and the probability of excursions outside chosen ranges.
Parameter estimation
Parameters can be estimated from glucose-tolerance tests, clamp experiments, meal tests, insulin measurements or continuous monitoring, depending on the model.
Parameters should be interpreted according to the experiment and equations from which they were estimated.
Identifiability
A complex glucose–insulin model may fit observations well while individual parameters remain poorly determined.
Before giving physiological meaning to fitted parameters, it is important to ask whether the available measurements can actually distinguish them.
From physiology to control
Glucose–insulin models also lead naturally to control theory. If insulin delivery \(u(t)\) can be adjusted using glucose measurements, the system becomes a feedback-control problem.
This mathematical structure underlies research on automated insulin-delivery systems.
A practical modelling workflow
Define whether the variables represent concentrations, amounts or deviations from baseline. Write glucose input and removal separately. Write insulin secretion and clearance separately. Decide whether delayed insulin action needs its own state variable. Add meal absorption or exogenous insulin when relevant. Check equilibria and stability. Fit the model to suitable glucose and insulin data, quantify uncertainty and test identifiability before interpreting fitted parameters physiologically.